When and When Not to Write with AI Assistance

AI synthesizing and AI writing and sort of “meeting note-taking” “action item listing” is incredibly valuable.

AI’s ability to distill is, in my opinion, superhuman, especially if you take into account the cost and the time.

However, if some writing / output needs your personal perspective, your actual human knowledge, understanding, somehow your world context, your consciousness, your internal emotions and feelings and nuance, then you’re gonna only get maybe 50% of the value of AI-assisted writing. In some cases, that could still be worth it. If it’s instant and basically free and 50% of the value is there and that’s good enough for your use case, go for it. I’m thinking like internal company memos or something.

However, a lot of writing is not actually done to convey raw information such as meeting notes or action items, but instead some writing is to try to impart some almost unknowable personal take, some summing up of your entire human context into a nuanced decision or a feeling one way or another. When you’re hedging or raising fears or doubling down even when you don’t have good reason to double down, but it’s just a gut feeling.

Maybe that’s a good way to put it. Anything that needs a gut feeling can’t come from an AI writer, because they don’t have any guts.

(is someone trying to map the neurons in the guts and make an AI model out of it? Maybe that’s what LLMs are missing - we have replicated the brain neurons but now we need to replicate the gut neurons and have them interact somehow)

So I think I want to split my writing output into two types.

Anything that is for just information purposes, use AI to synthesize and write and publish and label it as such and explain at the top, “Yes, this was synthesized by AI, but it is synthesizing my human content (meeting transcription, notes, etc.) I did read through it all; I endorse it all, but read it for it’s information and know that there may be missing nuance.”

Then other writing that needs my human perspective, my personal context, nuance, my gut feeling, I should transcribe with my voice, lightly edit by hand, maybe figure out a way to help AI clean up my writing but without changing any nuance… which might be hard. The problem here might be that your poor writing, even if it’s tactically poor writing, contains useful signal that gets sanded down by AI. And so you might want to convey some complex information that can’t be written down in a few sentences, but if you write a few bad sentences, you might trigger someone to think in the way that you were thinking when you tried to write it. And maybe AI can’t yet model that?

In this case, you have the ability to label your writing as human writing and maybe people will value it more, take it more seriously because you are staking your reputation on the fact that it’s not AI written.

A lot of this is because people don’t know how much to value AI writing on a case by case basis. In some contexts it’s totally fine and completely replaceable from human writing, and in other contexts the AI writing is less than 10% of the value of the potentially more poorly written human writing. And it’s unfair for the reader to try to determine what’s happening when they’re just looking at a random piece of writing. Not only is it unfair, but they’ll probably properly assume it’s not worth reading and skip it, because most of the time they’d be right.

So you as the writer and generator of the ideas, if you’re doing it in conjunction with AI or not, should preface upfront what you’re presenting.